Integrating hydrological data and hydrochemical data to reduce uncertainty in mountain aquifer modeling

Mountain systems are an important source of high-quality freshwater. Mountain aquifers, found at higher altitudes, are typically composed of fractured rocks and karst systems, while alluvial aquifers, located in valleys, are more permeable and often connected to mountain aquifers. This connection transfers large water volumes downstream, supporting urban water needs. Additionally, groundwater serves as a natural water storage, sustaining river flows during dry periods and helping mitigate extreme droughts.

Despite the pivotal importance of groundwater in mountain systems, monitoring efforts in such environments remain limited, which hinders the efficiency of groundwater models. Data on hydraulic properties and piezometric heads are typically scarce. As a consequence, groundwater modeling suffers intrinsically from equifinality and high parameter uncertainty. 

We analyzed the uncertainty associated with the flow and transport model of the lower Chiese valley in the Italian Alps. The valley hosts an aquifer exploited to provide the water needed for the local fishery industry. The fisheries are exposed to the risk of contamination from the upstream industrial activities.  We developed the flow and transport model of the aquifer, depending on the following parameters: the homogeneous hydraulic conductivity of the aquifer, the Chiese riverbed conductivity, and the mountain block recharge from the east and the west hillslopes. Datasets available include groundwater level measurements in 41 wells and piezometers, the chemical signature in 3 piezometers, and PFOS  concentration in 7 wells, representing sensitive points.  We computed the posterior parameters pdfs by means of  Markov Chain Monte Carlo with three levels of information.  At the first level, we used only piezometric data, then at the second level, we added chemical signatures at the observation wells and springs, and finally, at the third level, we added  PFOS  concentrations at 7 observation wells. 

Simulations showed that groundwater levels alone allow a small reduction of uncertainty, i.e., the posterior pdfs of the parameters differ slightly from the prior ones.  By incorporating chemical and contaminant concentrations into the model calibration, we observe a considerable reduction of model uncertainty, with posterior pdfs significantly different from the prior ones. In particular, the posterior pdf of the hydraulic conductivity is very narrow, with the most probable value of  10^-1 m/s, which is compatible with the prevalence of sandy/gravel material through the entire formation, as shown by the available well logs. On the other hand, the posterior pdfs of the mountain block recharge and riverbed conductivity are narrower than the a priori ones, but the reduction in the amplitude is smaller for that of the hydraulic conductivity. This indicates that the primary source of model uncertainty lies in the exchange fluxes among the aquifer, the mountain block, and the river, with the chemical composition of the water and pollutant concentrations being the most effective data for reducing this uncertainty. Piezometric heads alone introduce little constraints to these fluxes. The proposed procedure can be applied to other Alpine valleys of mountain regions, where important water resources are threatened by overexploitation and contamination due to anthropic activities.